Merchants running WeChat bots on Yuanqi have four days to export and migrate or their customer-service bot goes dark.
AI NEWS
Today’s AI News
Chinese and English are the same day. This page lists every item, each with the original link, the author, and the engagement.
Latest day
November 5, 2026
11-04 07:50 – 11-05 08:00 (UTC+8)
Top 3
Li Jiaqi's livestream sells ~100,000 Qwen Office monthly passes in 30 minutes at ¥0.01 (list ¥28)
AI subscriptions are now sold like consumer goods: penny deals plus top streamers to buy first-time users (inference).
Peec AI doubles revenue in six months to $20M ARR helping brands get found in AI search
Customers now ask AI where to buy—small brands should check how AI search describes them (inference).
All
Tencent shuts Yuanqi on Nov 9 and ByteDance halts new Coze enterprise sales Nov 30 as no-code agent platforms shrink
Merchants running WeChat bots on Yuanqi have four days to export and migrate or their customer-service bot goes dark.
CollapseTencent shuts Yuanqi on Nov 9 and ByteDance halts new Coze enterprise sales Nov 30 as no-code agent platforms shrink
Tencent's official notice says Yuanqi shuts down on Nov 9, 2026, and agents published to WeChat official accounts and customer service will stop replying; new agents and API keys were already frozen on Oct 29. Data can be exported in-app before shutdown and from an export console until Jan 9, 2027, when it is deleted. The official replacement is Tencent Cloud ADP: knowledge bases, workflows and widgets import directly, but agent settings and custom tools must be rebuilt. Separately, per Coze's notice as relayed by several users on X, Coze enterprise Standard and Flagship plans stop new sales on Nov 30, legacy Volcano Engine plans end Dec 31 with balances refunded, and export details are 'to be announced'.
Li Jiaqi's livestream sells ~100,000 Qwen Office monthly passes in 30 minutes at ¥0.01 (list ¥28)
AI subscriptions are now sold like consumer goods: penny deals plus top streamers to buy first-time users (inference).
CollapseLi Jiaqi's livestream sells ~100,000 Qwen Office monthly passes in 30 minutes at ¥0.01 (list ¥28)
On the evening of Nov 2, Alibaba VP and Qwen Office CEO Chen Yusen joined Li Jiaqi's Singles' Day livestream to sell a Qwen Office basic monthly pass listed at ¥28 for ¥0.01. Sina Tech and the Economic Observer report about 100,000 passes sold in half an hour, which media converted into roughly 2 trillion tokens of quota, not revenue. The pass includes 500 credits and is limited to first-time buyers. Beijing News, citing Tmall, says Taobao's token category GMV on the first sale day was up 450% month on month.
Peec AI doubles revenue in six months to $20M ARR helping brands get found in AI search
Customers now ask AI where to buy—small brands should check how AI search describes them (inference).
CollapsePeec AI doubles revenue in six months to $20M ARR helping brands get found in AI search
Berlin startup Peec AI said on Nov 2 it passed $20M ARR less than two years after launch, doubling from $10M in May 2026. It cites 4,000+ customers in 80+ countries, including Siemens, Hugo Boss and Sky, and a team of 100+ with a new New York GTM hub. The product helps marketers track and improve how brands are mentioned and recommended in AI search. Figures are company-reported and unaudited.
Report: Apple cuts some engineers’ daily AI spend cap $300→$200, maybe to $100 (Bloomberg-sourced)Unconfirmed
Even Big Tech meters tokens per seat—SMBs should set budgets and premium-model allowlists first.
CollapseReport: Apple cuts some engineers’ daily AI spend cap $300→$200, maybe to $100 (Bloomberg-sourced)
Wall St Engine/Techmeme citing Bloomberg’s Mark Gurman said Apple lowered some engineers’ daily AI token caps from about $300 to $200 last month, with cuts toward $100 under review; some non-engineering teams may fall from $100 to as little as $25, with cheaper or local models encouraged—even as FY26 R&D rose ~32% on infra/AI. Unconfirmed sourcing; no Apple official note.
CAICT and 20 firms including Huawei, Alibaba and Tencent sign an 8-point AI agent self-regulation pact
Agents that act or pay for users will likely need identity, authorization and audit trails; design for it now (inference).
CollapseCAICT and 20 firms including Huawei, Alibaba and Tencent sign an 8-point AI agent self-regulation pact
At the Wuzhen World Internet Conference on Nov 4, CAICT and 20 organisations including Tencent, Ant, Alibaba, Douyin, Huawei, Xiaomi and Baidu published an AI agent self-regulation pact (reported by The Paper, IT Home and Jiemian). Its eight articles cover verifiable layered agent identity, permission control, emergency cut-offs, and users' final say on important matters. Article 5 commits signatories to oppose using agents for algorithmic exploitation and fraud. It is industry self-regulation, not law, and carries no penalties.
China's Apple Intelligence enters beta with iOS 27.3: Siri, Writing Tools and Visual Intelligence run on Qwen, release in December
Qwen gets a system-level iPhone entry point; app and content teams should watch whether discovery shifts to Siri (inference).
CollapseChina's Apple Intelligence enters beta with iOS 27.3: Siri, Writing Tools and Visual Intelligence run on Qwen, release in December
IT Home and The Paper report that Apple pushed the first iOS 27.3 developer beta on Nov 4, starting the China Apple Intelligence beta on iPhone 15 Pro and later. Siri, Writing Tools and Visual Intelligence call Alibaba's Qwen, while visual 'search' uses Baidu image search, and results sync to the Qwen app. The public release is planned for December. Early user posts on X mention some queries bouncing to Baidu search and some prompts being blocked (anecdotal).
Skild AI passes $100M ARR ten months after first commercial deployment, 60+ paying customers
Robots are now sold on invoices, not demos—warehouses can ask for per-task pricing and small pilots (inference).
CollapseSkild AI passes $100M ARR ten months after first commercial deployment, 60+ paying customers
AI Magazine reported on Nov 4 that robot foundation-model company Skild AI passed $100M ARR (as of Sept 2026) ten months after first commercial deployment, with 60+ paying customers and partners including ABB, NVIDIA and Foxconn; about $50M has been recognized since deployments began. Roughly 90% of revenue comes from manipulation tasks and 10% from mobility. Company figures relayed by media, unaudited.
NiCE Q3: AI ARR $392M (+46%); using AI to streamline, cutting ~5% of workforce
Selling AI for growth while using AI to trim staff—call-center work feels it first; SMBs can price out AI for support (inference).
CollapseNiCE Q3: AI ARR $392M (+46%); using AI to streamline, cutting ~5% of workforce
Customer-service software firm NiCE reported Q3 on Nov 4: revenue and EPS above the top of guidance, AI ARR of $392M (+46% YoY, 16% of cloud revenue), and a raised full-year EPS outlook. It also said it will use AI to cut management layers and reduce global headcount by about 5%, mainly in product, go-to-market, support and services, with $15–20M restructuring charges mostly in Q4. Figures per the company release and call.
Moonshot opens Kimi K3.1 Pro weights: 2.8T parameters, API at $2/$10 per million input/output tokens
Another top-tier Chinese model you can self-host commercially; most small teams will still find the API cheaper (inference).
CollapseMoonshot opens Kimi K3.1 Pro weights: 2.8T parameters, API at $2/$10 per million input/output tokens
Moonshot announced on X at midnight Nov 5 (UTC+8) that Kimi K3.1 Pro weights are public on Hugging Face. The model card lists a MoE with 2.8T total and 104B active parameters, a 1M-token context and about 1.56TB of weights. The API id is kimi-k3.1-pro; pricing cited by dev.to is $2 input, $10 output and $0.20 cached input per million tokens. The licence allows commercial use, but model-as-a-service firms above $20M annual revenue need a separate deal, and products above 100M MAU must display Kimi.
Elon Musk: Grok Bot will research purchases, order, track and return what you don’t want
Vision posts aren’t SLAs—merchants need agent spend caps and human-confirm switches first.
CollapseElon Musk: Grok Bot will research purchases, order, track and return what you don’t want
Elon Musk posted that Grok Bot will research purchase ideas, bring options, place orders, track shipments and return unwanted items. Product-vision post; regional availability, payment risk controls and merchant refusals weren’t detailed—alongside Uber’s “4 bookings” datapoint, agent shopping remains early.
MercadoLibre: will not feed inventory to third-party AI shopping platforms for now
A major LatAm retailer cutting agents off means brands must clear inventory rights before Shop MCP/agent deals.
CollapseMercadoLibre: will not feed inventory to third-party AI shopping platforms for now
Wall St Engine reported MercadoLibre ($MELI) said it will not provide inventory to any third-party AI shopping platform for now, keeping listings out of agentic AI shopping. A clear platform refusal of agent-led discovery/checkout; scope and duration follow company statements.
Xiaomi humanoid robot works 8h05m straight in its car factory, covering a full day at one sorting station
Humanoids moved from demos to a full shift, but without success-rate or cost data, manufacturers should wait (inference).
CollapseXiaomi humanoid robot works 8h05m straight in its car factory, covering a full day at one sorting station
Xiaomi released a video on Nov 3 of its full-size humanoid completing 8h05m of continuous work at the console side-cover sorting station in its car plant, meeting the station's full-day demand with self-correction and line coordination. STAR Market Daily says the robot charged twice, at 3h02m and 6h03m, and a wheeled robot picked up one dropped part. It is Xiaomi's third public milestone this year, after a 90.2% success rate in March and 98% in July. Electrek notes Xiaomi did not disclose the 8-hour success rate or the number of human interventions.
Greencart: in-app agent Kapok placed ~250 Amazon orders in a week—only after user yes
“Buy only after yes” is table stakes—make second confirmation non-optional in agent integrations.
CollapseGreencart: in-app agent Kapok placed ~250 Amazon orders in a week—only after user yes
Greencart claimed its in-app AI shopping agent Kapok searched, compared and submitted about 250 Amazon orders in a week, with nothing purchased without a user yes. Startup marketing metrics; success/cancel rates undisclosed—treat cautiously.
TechCrunch: Uber says only 4 people booked a ride via an AI agent in its busiest week last quarter
Don’t assume agents already drive demand—pilot with measurable traffic before revenue share talks.
CollapseTechCrunch: Uber says only 4 people booked a ride via an AI agent in its busiest week last quarter
TechCrunch reported Uber said only four people booked a ride using an AI agent during its busiest week last quarter—underscoring how thin third-party agent checkout still is versus hype. Exact definitions follow Uber’s wording.
dotey, citing Reuters: Google offers voluntary exit packages in 5 countries, up to 12 months' payUnconfirmed
Big tech is cutting costs with voluntary exits rather than layoffs, a softer playbook managers can study (inference).
Collapsedotey, citing Reuters: Google offers voluntary exit packages in 5 countries, up to 12 months' pay
dotey relayed a Reuters report that Google is offering voluntary exit packages in five countries, worth up to 12 months' salary. The post has about 29,000 views. Which countries and teams are covered, and how this ties to AI spending, should be checked against Reuters. It is another sign of big tech raising AI investment while trimming headcount costs.
OpenAI launches ChatGPT Images 2.6: GIFs + reference images; API Flare/Sunburst ~50% faster
Faster image loops help SMBs, but acceptance should track brand consistency across edit rounds—not first-image latency alone.
CollapseOpenAI launches ChatGPT Images 2.6: GIFs + reference images; API Flare/Sunburst ~50% faster
On Nov 4 OpenAI launched ChatGPT Images 2.6 across ChatGPT, Codex and the API—animated GIFs from prompts or photos, reusable reference images, and ~50% faster generation. The API adds GPT-Image-2.6 Flare (as little as ~6s at low quality) and higher-fidelity Sunburst, plus image markup, SVG export and a complexity control. OpenAI says about 4 billion images are created weekly across products. Features and pricing follow the product page and system card.
OpenAI’s Greg Brockman: Pierre Computer joins OpenAI to advance Codex collaborative coding
Agentic IDE M&A is speeding up—SMB buyers should ask about lock-in and data residency.
CollapseOpenAI’s Greg Brockman: Pierre Computer joins OpenAI to advance Codex collaborative coding
OpenAI president Greg Brockman welcomed the Pierre Computer team, saying their open-source developer tools already power parts of Codex and that they will build the future of collaborative software with AI. Andrew Curran called it unusual talent moving from Anthropic’s orbit. Deal/product details await further disclosure.
dotey highlights Cicada, an open-source native macOS live-translation meeting app (135 bookmarks vs 84 likes)
A free option for small teams meeting overseas clients before paying for a translation subscription (inference).
Collapsedotey highlights Cicada, an open-source native macOS live-translation meeting app (135 bookmarks vs 84 likes)
dotey introduced Cicada, an open-source, fully native macOS app for live speech translation and AI meeting notes. The post drew 135 bookmarks against 84 likes, a sign many people saved it to try. Exact features, supported models and languages should be checked in the post and project page. It suits small teams that meet overseas clients.
The Information: voters passed 43 of 44 local measures restricting data centers; Anthropic-backed PAC eyes 2028
Community vetoes are hardening—cloud/compute buyers need multi-region fallbacks, not one campus bet.
CollapseThe Information: voters passed 43 of 44 local measures restricting data centers; Anthropic-backed PAC eyes 2028
The Information reported voters approved 43 of 44 local ballot measures restricting data centers, many with >70% support. Brad Carson of the Anthropic-backed Public First super-PAC network said the issue will likely dominate the 2028 cycle. Election results and political commentary; federal path remains uncertain.
OpenAI: autonomous cyber pipeline validated patches for 948 Linux LPEs at ~$1.2M API cost
If finding beats fixing on cost, enterprises and OSS maintainers need automated patch review as routine—not a bounty backlog.
CollapseOpenAI: autonomous cyber pipeline validated patches for 948 Linux LPEs at ~$1.2M API cost
On Nov 4 OpenAI said its security team ran an autonomous discovery–validation–exploit–patch pipeline with an upcoming cyber model: one mid-October run dynamically reproduced ~1,724 vulns and produced 948 working Linux local privilege-escalation exploits with candidate patches on a stock desktop config—about $1.213M in API credits (~$1,279 per LPE). It claims at least 10 novel exploitation techniques in a sample, with human review before reports to Linux maintainers and Patch the Planet partners. Research announcement; upstream merge status is separate. What it means for ordinary businesses (inference): If finding beats fixing on cost, enterprises and OSS maintainers need automated patch review as routine—not a bounty backlog.